Discussion Apple Silicon SoC thread

[Eug]

M1
5 nm
Unified memory architecture - LP-DDR4
16 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 12 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache
(Apple claims the 4 high-effiency cores alone perform like a dual-core Intel MacBook Air)

8-core iGPU (but there is a 7-core variant, likely with one inactive core)
128 execution units
Up to 24576 concurrent threads
2.6 Teraflops
82 Gigatexels/s
41 gigapixels/s

16-core neural engine
Secure Enclave
USB 4

Products:
$999 ($899 edu) 13" MacBook Air (fanless) - 18 hour video playback battery life
$699 Mac mini (with fan)
$1299 ($1199 edu) 13" MacBook Pro (with fan) - 20 hour video playback battery life

Memory options 8 GB and 16 GB. No 32 GB option (unless you go Intel).

It should be noted that the M1 chip in these three Macs is the same (aside from GPU core number). Basically, Apple is taking the same approach which these chips as they do the iPhones and iPads. Just one SKU (excluding the X variants), which is the same across all iDevices (aside from maybe slight clock speed differences occasionally).

EDIT:



M1 Pro 8-core CPU (6+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 14-core GPU
M1 Pro 10-core CPU (8+2), 16-core GPU
M1 Max 10-core CPU (8+2), 24-core GPU
M1 Max 10-core CPU (8+2), 32-core GPU

M1 Pro and M1 Max discussion here:

Page 78 - Discussion - Apple Silicon SoC thread

Page 78 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M1 Ultra discussion here:

Page 109 - Discussion - Apple Silicon SoC thread

Page 109 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M2 discussion here:

Page 127 - Discussion - Apple Silicon SoC thread

Page 127 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

Second Generation 5 nm
Unified memory architecture - LPDDR5, up to 24 GB and 100 GB/s
20 billion transistors

8-core CPU

4 high-performance cores
192 KB instruction cache
128 KB data cache
Shared 16 MB L2 cache

4 high-efficiency cores
128 KB instruction cache
64 KB data cache
Shared 4 MB L2 cache

10-core iGPU (but there is an 8-core variant)
3.6 Teraflops

16-core neural engine
Secure Enclave
USB 4

Hardware acceleration for 8K h.264, h.264, ProRes

M3 Family discussion here:

Page 215 - Discussion - Apple Silicon SoC thread

Page 215 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

M4 Family discussion here:

Page 263 - Discussion - Apple Silicon SoC thread

Page 263 - Seeking answers? Join the AnandTech community: where nearly half-a-million members share solutions and discuss the latest tech.

forums.anandtech.com

 

[Eug]

A17 Pro SoC
---
19 billion transistors, N3 (EUV)

I just realized that's almost 20% more transistors than M1 at 16 billion, and is >60% more transistors than A14 upon which M1 is based) at 11.8 billion.

The first Apple-utilized CPU I owned was the 6502, and it had just 3510 transistors. (Or technically 4528.)

P.S. That initial post on A17 Pro was the first post of page 200 in this thread, and we are now over 5000 posts.

 

[A///]

I just realized that's almost 20% more transistors than M1 at 16 billion, and is >60% more transistors than A14 upon which M1 is based) at 11.8 billion.

The first Apple-utilized CPU I owned was the 6502, and it had just 3510 transistors. (Or technically 4528.)

P.S. That initial post on A17 Pro was the first post of page 200 in this thread, and we are now over 5000 posts.

Early Apple employee?

 

[Eug]

Early Apple employee?

No, I’m just old. My first desktop was a clone of the Apple ][+, back in the early 80s. The clones all ran the same 6502.

 

[A///]

No, I’m just old. My first desktop was a clone of the Apple ][+, back in the early 80s. The clones all ran the same 6502 that Apple did.

Ah. Retiring in a few years myself and then spending the rest of my days doing nothing but relaxing. Never had a thought to keep old computers and gizmos from back then and instead bought and collected typewriters which I then sent off to various places. About 20 years later I realised there's a market for them and they fetch eye watering numbers. Drat.

 

[Doug S]

Yeah they highlighted pretty heavily that this GPU core was basically the biggest redesign they have ever done. Also that 20% figure was peak performance gain, they never mentioned anything about sustained performance boost, so it could end up being lower than 20% in games.

Yep doesn't make sense. Now sure if that major redesign was mostly about the hardware ray tracing I could understand that's where they concentrated their efforts, but it would be silly to go out of their way to avoid making any improvements in the GPU's non ray tracing performance. To get only 20% from a 6th core would mean that no only did they not make any improvements to its "traditional" GPU performance, they would have to have managed to get zilch out of N3. Not even a 1 MHz clock boost!

 

[A///]

Wasn't it discussed a while back the new A17 Pro chip wasn't a huge overhaul design?

 

[Eug]

Yep doesn't make sense. Now sure if that major redesign was mostly about the hardware ray tracing I could understand that's where they concentrated their efforts, but it would be silly to go out of their way to avoid making any improvements in the GPU's non ray tracing performance. To get only 20% from a 6th core would mean that no only did they not make any improvements to its "traditional" GPU performance, they would have to have managed to get zilch out of N3. Not even a 1 MHz clock boost!

Maybe they emphasized power efficiency, while doubling the neural engine speed (35 vs 17 trillion ops), and adding several hardware accelerators including AV1, computional photography support for the new periscope lens etc., plus GPU features, the extra GPU core, and a bit of CPU boost. It should be noted that this is in the context of rumours of mediocre yields on N3B, on a chip with an increase in the number of transistors by almost 20%.

Also, I wonder if they really will switch from N3B to N3E next year.

 

[LightningZ71]

No, I’m just old. My first desktop was a clone of the Apple ][+, back in the early 80s. The clones all ran the same 6502.

Laser 128?

As to the A17 (and likely related M3) I feel that they are where they need to be for this generation. They will still have a modest amount of total performance/power leadership in their chosen markets and it enables them to continue onwards to achieving more on the next refresh. We should also keep in mind that there may be a more notable difference between the A17 and the M3 than the one between A16 and M2. A17 is completely bent towards being mobile. M3 will be focused more on laptop and scaling to desktop. M3 offers a degree of flexibility that previous nodes didn't have with respect to tailoring how the transistors are configured and Apple could take a different approach using those changes.

 

[Mopetar]

Well, one piece of good news (besides the GPU stuff) is it finally got a hardware AV1 decoder. That means M3 will also have a hardware AV1 decoder.

Is that even a big deal for most people?

Apple is such a large install base that any big company is just serving Apple users H264/5 encoded media and probably will continue to for some time since most Apple devices won't have an AV1 hardware decoder for some time.

 

[Eug]

Laser 128?

Zeus. I think it may have been the Zeus 2001.

It was an Apple ][+ clone which I modded for lower case support, and added a 16K card. I also added a CP/M card (among other things), although I had zero use for it CP/M.

And then my mom gave it away without my permission when I was away in university. I was so mad. The good news is she gave it to a computer science professor who happened to specialize in the 6502... which by that time was completely out of date, but the guy was from China that was decades behind technologically at that time. I remember as a kid visiting a Chinese university in the 80s and the grad students were learning to program on punch tape machines.

Is that even a big deal for most people?

Apple is such a large install base that any big company is just serving Apple users H264/5 encoded media and probably will continue to for some time since most Apple devices won't have an AV1 hardware decoder for some time.

4Kp60 support has been hit and miss in YouTube on the Mac, and even if there is AV1 software support, it kills the battery. Presumably with Apple embracing AV1 finally, it will mean consistent 4Kp60 AV1 support (and 8K support) with long battery life.

Is it critical for most people? No. Is it nice to have for some people? Yes!

Put it this way: Back in the day I waited a long time before replacing my 2009 13" MacBook Pro and my 2010 27" iMac. I finally bit the bullet in 2017 with a 12" MacBook and another 27" iMac, partially because they have hardware support for h.265 HEVC with their Kaby Lake CPUs. These days I don't think most enthusiasts would be happy with a machine that didn't have hardware h.265 HEVC support. Similarly, in a few years I don't think most enthusiasts would be happy with a machine that didn't have hardware AV1 support.

For the Sony Nature HEVC demo video, a 2016 27" iMac Core i7 cannot decode it cleanly in software, even with 100% CPU usage. My 2017 27" iMac Core i5 can do it with 10% CPU usage, and my 2017 12" MacBook can do it with 25% CPU usage.

The results won't be as dramatic for AV1 on Apple Silicon, but for mobile machines it will still mean a significant improvement in battery life whenever AV1 decoding is involved.

 

[Ajay]

Every single Apple chip uses piles of LPDDR.

128Gb LPPDR5 DRAM chips will be HVM soon. I think it was Ming or Gurman who suggested increased RAM coming for the Studio and Mac Pro. Guess we'll see.

 

[Doug S]

Is that even a big deal for most people?

Apple is such a large install base that any big company is just serving Apple users H264/5 encoded media and probably will continue to for some time since most Apple devices won't have an AV1 hardware decoder for some time.

Yeah I don't see why everyone is so excited about this. Of all the things Apple did or didn't do, AV1 would be at the bottom of the list for me. It would make no difference to me if they never did it, yet people are acting like this is the biggest upgrade coming with iPhone 15. I don't get why people appear to care so much about something so trivial.

 

[adroc_thurston]

128Gb LPPDR5 DRAM chips will be HVM soon

No, 32Gbit QDPs are some time away.

 

[Ajay]

No, 32Gbit QDPs are some time away.

What is a QDP, google is failing me?

 

[Doug S]

128Gb LPPDR5 DRAM chips will be HVM soon. I think it was Ming or Gurman who suggested increased RAM coming for the Studio and Mac Pro. Guess we'll see.

Those aren't "chips", they are modules. I assume you're talking about a cell phone config with four 32 Gb x16 DRAMs in an LPDDR5 module? DRAM chips are currently 24 Gb going to 32 Gb next year.

Since Apple uses only two different configs for the 128 bit wide stacks used in Apple Silicon Macs, that would mean the M2 Mac Pro going from the current 384 GB max to 512 GB. But if they used the maximum number of chips possible in those stacks, they could hit 3 TB with the 24 Gb DRAMs and 4 TB with the upcoming 32 Gb DRAMs. Obviously the market for a 4 TB Mac Studio could be counted on one hand, or probably no hands, but they must feel even the next step up to allow 1 TB isn't large enough to bother with.

 

[adroc_thurston]

What is a QDP, google is failing me?

Quad die package.

 

[Eug]

Yeah I don't see why everyone is so excited about this. Of all the things Apple did or didn't do, AV1 would be at the bottom of the list for me. It would make no difference to me if they never did it, yet people are acting like this is the biggest upgrade coming with iPhone 15. I don't get why people appear to care so much about something so trivial.

For many people hardware video acceleration is much, much more important than raw CPU power or ray tracing or whatever.

As mentioned previously, one of the most important CPU feature upgrades for me in my last laptop purchase was hardware h.265 acceleration. AV1 is the next generation.

Mind you, hardware h.265 acceleration was a bigger jump than I expect hardware AV1 acceleration to be for Apple. Apple built much of its video ecosystem around h.265, but I'm not convinced Apple will do the same with AV1. Nonetheless, AV1 is becoming a main standard for video distribution so it's going to be very important going forward.

 

[A///]

Nonetheless, AV1 is becoming a main standard for video distribution so it's going to be very important going forward.

A very slow train at that. Fast forward 6-7 years from now people will be on 10th or 12th gen intel processors or whatever AMD processors and with the latest video card wondering why their performance isn't as good as others. sadly there's too many people on the internet who are running Intel processors from the mid 2010s with a 4090 and wondering and complaining why performance on their system sucks compared to others.

Av1 is possibly the most exciting video file format in a long time. I was dully impressed with how clear h.265 was at a fraction of a file size of h.264 when I first happened across video files.

I'm even more excited to get my iphone 15 pro. It'll be a decent upgrade from my current iPhone that's several years old now. I was considering holding onto it and having the battery replaced as its begun to show its age but I'll take the new phone instead. I'd like to see Android catch up to Apple in terms of quality one day but I doubt it'll ever come to fruition.

 

[Ajay]

Those aren't "chips", they are modules. I assume you're talking about a cell phone config with four 32 Gb x16 DRAMs in an LPDDR5 module? DRAM chips are currently 24 Gb going to 32 Gb next year.

Since Apple uses only two different configs for the 128 bit wide stacks used in Apple Silicon Macs, that would mean the M2 Mac Pro going from the current 384 GB max to 512 GB. But if they used the maximum number of chips possible in those stacks, they could hit 3 TB with the 24 Gb DRAMs and 4 TB with the upcoming 32 Gb DRAMs. Obviously the market for a 4 TB Mac Studio could be counted on one hand, or probably no hands, but they must feel even the next step up to allow 1 TB isn't large enough to bother with.

Thanks! I misunderstood Micron news article. Yeah, seems to me boosting the Mac Pro to 512 GB makes sense, it's a niche product and expensive, so why not go all in needed or not.

 

[Ajay]

Quad die package.
View attachment 85768

Thanks, found the semiengineering.com article. Learning something new today. Guess I owe myself a cookie

 

[A///]

Thanks, found the semiengineering.com article. Learning something new today. Guess I owe myself a cookie

Throw me a biscuit too. I've ate all mine.

Will have to read that article. Got a link?

 

[adroc_thurston]

Thanks, found the semiengineering.com article. Learning something new today. Guess I owe myself a cookie

There are also ODPs (Octal Die Packages) which run LPDDR in that weird 8b subchannel mode.
Grace uses them.
And then there's Apple's custom ODP without subchanneling for Max/Ultra parts (128b per package things).

 

[Ajay]

Throw me a biscuit too. I've ate all mine.

Will have to read that article. Got a link?

Thin Quad Die Package (QDP) Development

A new packaging format provides extreme silicon area density versus package volume.

semiengineering.com

 

[A///]

Very interesting read. Wish it were longer. Thanks @Ajay

 

[Henry swagger]

Apple’s A17 Bionic Barely Beats The A16 Bionic In New Multi-Core Leak, Only Manages A 10 Percent Improvement In Single-Core Result

The A17 Pro is a 3nm SoC, but it fails to deliver a substantial single and multi-core performance lead over A16 Bionic, according to new leak

wccftech.com

Looks like tsmc n3 node is bad or arm isa has it a brick wall.. even apple said a17 has a wider decode than a16 🙄